Counting Cats in Zanzibar Rotating Header Image

I don’t believe you want to do that Dave…

Well it would appear possibly, arguably, a computer at the Royal Society in London has passed the Turing test.

Read the whole thing. It is interesting. Alas I seem unable to copy and paste from the Guardian otherwise I’d dissect this because I am less than impressed. It would appear they haven’t released the transcripts. And it was impersonating a 13 year old boy. All very fishy. Certainly it ain’t as tough a test as proposed by Kurzweil. This had to be believed by 30% and got 33%. The Kurzweil test is much more rigorous.

Oh and Prof. Kevin Warwick was involved. Hmm…

My fave comment though on the Graun is this (I seem to be able to copy those)…

For a moment, let’s just forget whether and why some computer might pass the test and what that might mean. Suppose instead you wanted to decide whether a human is intelligent… What criteria would you apply? What rigorous and material or empirical definition could you come up with for “intelligent”? Or for “thought”, “objective”, “emotion” or any other noun relating to individuals inner lives for that matter?

Of course there’s no real definition for “intelligent” that doesn’t rely on other abstract nouns, e.g. if you decide it’s “problem solving ability” then you only shift the question along to “what’s a problem, then?”.

But we all agree as a linguistic convention that there is such a thing as intelligence and that humans possess it. But if that’s true and a computer and successfully disguise itself in some open-ended way as a human then we’ve no grounds for denying the title of “intelligent” to the machine.

You may still deny that this has any metaphysical significance. On the other hand, you can’t deny that in that hypothetical the computer has transcended your ability to distinguish it from other entities you agree to be intelligent. That makes the machine categorically distinct from all others in history, at least from your perspective and is surely a significant fact in itself. Without the Turing test, you’d be stuck in a quagmire — what the test does is isolate this significant observation from all metaphysical or linguistic confusion, reducing the matter to observable behaviour.

In the end, it’s a definition of intelligence. Do you have a better one?

No, I don’t but I have never felt sure about the Turing test in general – and yes I have read a lot about it. Does it have agency? Does it have imagination? Can it make mistakes? Mistakes are important for creativity. They seem to me to link tightly with creativity. I have for a long time thought it is probably in principle to compare genuine thought with the perfection of computers in a way almost analogous to quantum complementarity. I have no idea why I feel this except I feel it which perhaps is the point. I also tend to think the Turing test is just too instrumentalist. It in a sense doesn’t get to the heart of consciousness. It’s Searle’s Chinese Room. It is sort of a search for pure empirical proof without theory.

I never trust pure empiricism without theory. I think that might have been Eddington but I can’t track it down and I am writing in a rush. There is a saying (a joke really) in the AI biz about whether you can take it apart with itself? And that is the problem. Can we really understand ourselves properly, scientifically? I feel it is an impossible task and more to the point pointless. Shakespeare couldn’t predict comets arriving but he knew humanity better than any trick-cyclist. Just look at Freud. Or Kinsey or any of those preverts. I’ll believe in the Turing test when it can explain why I love Donne’s 20th elegy but can’t stand Tennyson’s romantic musings. I’ll leave the last word to Albert Einstein…

It would be possible to describe everything scientifically, but it would make no sense; it would be without meaning, as if you described a Beethoven symphony as a variation of wave pressure.

16 Comments

  1. john in cheshire says:

    Why hasn’t this computer been set an IQ test? Wouldn’t that be the start of deciding if it’s intelligent in some way? Or were they actually trying to demonstrate something else; replicant human behaviour or something; and not intelligence per se?

  2. NickM says:

    Interesting point I hadn’t considered JIC but I don’t hold much with IQ tests either.

  3. Paul Marks says:

    A computer person (an agent) – as with the computer on the Moon in Heinlein’s “The Moon Is A Harsh Mistress”.

    I doubt it has happened yet – but it may happen soon (I just do not know).

    Would such a computer have a “soul” – in the Aristotelian sense YES.

    But in the religious sense?

    Well God (if He exists) could endow the computer with a soul on the moment of its sentience.

  4. Mr Ed says:

    How gormless was the 13-year old ‘control’ used in the experiment as a positive control? What do you mean there was not one? What sort of scientist does an experiment without positive and negative controls if they can be devised?

    How gormless were the staff testing the computer?

    Did anyone ask it if it was a computer?

    Did they get Gary Glitter in for verification?

    Note the swipe in the article at Mr Turing’s work not being celebrated after the war.

    “Instead of being hailed a hero, Turing was persecuted for his homosexuality.”

    He was prosecuted for what was a criminal offence at the time, and his work was most secret (although the Soviets appear to have had plenty of sources in Bletchley Park, which may be why it is so popular with the Left).

  5. Sam Duncan says:

    Interesting that JiC brings up IQ tests because I’ve always thought the Turing test is very similar to them, in that all it tests for is ability to pass the Turing test. In other words, it doesn’t test for actual, existing, artificial intelligence, but the ability to fool a human operator for long enough to believe he’s talking to an intelligence. Which is, don’t get me wrong, very hard and very cool, but AI types have always held it to amount the same thing – if it looks like a duck and quacks like a duck – and I’m not so sure.

    I think your Guardian commentator, taking the traditional AI view, misses the point; even if you could create an AI that could fool 99% of people 99% of the time, that tiny gap must still mean it’s not really intelligent. That 1% can still occasionally see behind the veil. What about 99.999999% of people, 99.999999% of the time? Good enough to be getting on with, perhaps, but still demonstrably not a genuine intelligence. So even if it reaches 100%, you still know that it’s merely a convincing facade for a fake, a very fancy command shell for a perfectly ordinary, thick as several short planks, computer.

    Which brings us to the trickiness of that word “artificial”: does it mean the real thing, created by humans rather than arising spontaneously, like artifical diamonds, or does it simply mean “fake”, “pretend”, “not-real”, like artificial leather? Is AI diamonds or leather? There is research into the former, with neural networks and suchlike, but it’s nowhere even close to making an assault on the Turing test yet. Attempts to pass it always seem to be the latter. Which, in a perfect Platonic world, would be indistinguishable from the real thing, but this isn’t that world.

    So this bunch of code has passed (as Nick says, a very limited implementation of) the Turing test, but does anyone seriously believe that it’s a genuinely intelligent being, on a par even with a dog or a fish? It’s just as much a parlour trick as the chess-playing Turk, albeit vastly more sophisticated. (And, again, to be clear, very, very cool, but let’s not be fooled.)

  6. cosmic says:

    “In the end, it’s a definition of intelligence. Do you have a better one?”

    Agrees with me.

  7. NickM says:

    Mr Ed,
    A good point. If it had been a thirty year old then that makes part of a difference.

    Sam,
    My point in quoting the Graun commentator was that the definition thing matters. The Turing test like IQ tests are essentially definitions. They are arbitary in much the same way perhaps that the rules of a sport are.

    The “artificial point” you raise is important. I think if you devise something that seems smart at some level it is a thing not a being. But if you create something that learns and grows then we potentially have a sentience (or soul as Paul puts it). But how useful is it? The reason this laptop is useful to me is it is a machine and short of Windows going off on one it does as told in a predictable manner. It doesn’t have a personality. It never slacks off or tells me to sod off or decides to do it’s own thing. In short it is never going to decide to sing “Daisy, Daisy” or go Roy Batty.

  8. RAB says:

    If the machine had finally said… fuck this, I’m bored and not playing anymore. I’m off for a pint and a fag…. Then we’d know we had created artificial intelligence. ;-)

  9. John Galt says:

    I think that Sam hits closest to the bullseye. The Turing Test was originally and primarily about building an artificial intelligence (i.e. independent thought, agency, etc.) and the measure of this artificial intelligence is whether a human could be convinced they were talking to a human because of its artificial intelligence.

    What we seem to be doing though is creating systems which mimic stereotypical behaviours and then asking the human, how good was the simulacrum?

  10. CountingCats says:

    Sam,

    “if it looks like a duck and quacks like a duck” it could still, nonetheless, be just a computer simulation.

  11. Julie near Chicago says:

    Nick, thus:

    There is a saying (a joke really) in the AI biz about whether you can take it apart with itself? And that is the problem. Can we really understand ourselves properly, scientifically?

    Kronecker: “It is terribly difficult to see a field, from within the field.”

    (Disclosure: Quoted from what I laughingly call my memory.)

  12. Mr Ed says:

    What if the test were reversed, and a computer invited to distinguish the responses to it of a human from those of a computer to computer?

  13. Julie near Chicago says:

    The Computer, the one being tested, puts the question:

    Define the Universe and give two examples.

    The computer being queried goes into an infinite loop.

    Only the human answers:

    So why are you asking me? I’m only a University of Chicago first-year.

    But The Computer does not respond. It has ascended to the Astral Plane, where it can contemplate in eternal silence the beauty and mystery of its logic gates.

  14. endivior says:

    I dunno if anyone else here remembers ELIZA. I tried it out back in the 00s and remember thinking “well, it doesn’t seem all that intelligent, but it could certainly have fooled me into thinking it was a real psychotherapist.”

  15. Uncle Gus says:

    Mr Ed; Turing was prosecuted under an unjust law that more influential men routinely evaded. And because of that, his groundbreaking work was more or less ignored for more than a generation.

    And I’m d***ed if I can detect any left-wing bias in the general public hero-worship of Bletchley Park.

  16. Julie near Chicago says:

    UPDATE: CORRECTION.

    This stinks. I had thought in the first place that the words I attributed to Prof. Kronecker above were in fact spoken by psychologist Stanley Milgram (of the famous punishment experiments).

    At some point I Looked It Up and discovered it was actually from Prof. K. Or so I have remembered it, and have been telling it, for several years now.

    However, for some reason tonight I felt this uncontrollable urge to verify the attribution. And this time, I find a source which claims it was said by the dude Emerson, and in fact rather less elegantly:

    The field cannot well be seen from within the field. — Emerson

Leave a Reply

%d bloggers like this: